[Vision/Object Detection]What is the best model for Face Mask Detection? - 코드구현
목적
- 마스크 착용 여부를 가장 정확하게 탐지하는 모델을 찾아보자.
- 7개의 모델을 비교하고자 한다.
- 설명 슬라이드 및 논문 페이퍼(링크)
Configuration
- Load Kaggle Dataset
- Import Libraries
#코랩과 캐글 연동 및 다운로드
!pip install kaggle
from google.colab import files
files.upload()
!mkdir -p ~/.kaggle
!cp kaggle.json ~/.kaggle/
!chmod 600 ~/.kaggle/kaggle.json
!kaggle datasets download -d ashishjangra27/face-mask-12k-images-dataset
!unzip face-mask-12k-images-dataset.zip
!ls
Requirement already satisfied: kaggle in /usr/local/lib/python3.7/dist-packages (1.5.12)
Requirement already satisfied: certifi in /usr/local/lib/python3.7/dist-packages (from kaggle) (2020.12.5)
Requirement already satisfied: python-dateutil in /usr/local/lib/python3.7/dist-packages (from kaggle) (2.8.1)
Requirement already satisfied: tqdm in /usr/local/lib/python3.7/dist-packages (from kaggle) (4.41.1)
Requirement already satisfied: six>=1.10 in /usr/local/lib/python3.7/dist-packages (from kaggle) (1.15.0)
Requirement already satisfied: python-slugify in /usr/local/lib/python3.7/dist-packages (from kaggle) (5.0.2)
Requirement already satisfied: requests in /usr/local/lib/python3.7/dist-packages (from kaggle) (2.23.0)
Requirement already satisfied: urllib3 in /usr/local/lib/python3.7/dist-packages (from kaggle) (1.24.3)
Requirement already satisfied: text-unidecode>=1.3 in /usr/local/lib/python3.7/dist-packages (from python-slugify->kaggle) (1.3)
Requirement already satisfied: idna<3,>=2.5 in /usr/local/lib/python3.7/dist-packages (from requests->kaggle) (2.10)
Requirement already satisfied: chardet<4,>=3.0.2 in /usr/local/lib/python3.7/dist-packages (from requests->kaggle) (3.0.4)
....
[1;30;43m스트리밍 출력 내용이 길어서 마지막 5000줄이 삭제되었습니다.[0m
inflating: Face Mask Dataset/Train/WithoutMask/1861.png
inflating: Face Mask Dataset/Train/WithoutMask/1862.png
inflating: Face Mask Dataset/Train/WithoutMask/1863.png
inflating: Face Mask Dataset/Train/WithoutMask/1864.png
inflating: Face Mask Dataset/Train/WithoutMask/1865.png
inflating: Face Mask Dataset/Train/WithoutMask/1866.png
....
inflating: Face Mask Dataset/Validation/WithoutMask/964.png
inflating: Face Mask Dataset/Validation/WithoutMask/965.png
inflating: Face Mask Dataset/Validation/WithoutMask/97.png
inflating: Face Mask Dataset/Validation/WithoutMask/970.png
inflating: Face Mask Dataset/Validation/WithoutMask/971.png
inflating: Face Mask Dataset/Validation/WithoutMask/974.png
face-mask-12k-images-dataset.zip kaggle.json
'Face Mask Dataset' sample_data
!kaggle datasets download -d andrewmvd/face-mask-detection
Downloading face-mask-detection.zip to /content
99% 393M/398M [00:02<00:00, 211MB/s]
100% 398M/398M [00:02<00:00, 182MB/s]
!unzip face-mask-detection.zip
Archive: face-mask-detection.zip
inflating: annotations/maksssksksss0.xml
inflating: annotations/maksssksksss1.xml
inflating: annotations/maksssksksss10.xml
inflating: annotations/maksssksksss100.xml
...
inflating: images/maksssksksss0.png
inflating: images/maksssksksss1.png
inflating: images/maksssksksss10.png
inflating: images/maksssksksss100.png
inflating: images/maksssksksss101.png
inflating: images/maksssksksss102.png
! kaggle datasets download -d niharika41298/withwithout-mask
Downloading withwithout-mask.zip to /content
27% 9.00M/33.5M [00:00<00:01, 23.8MB/s]
100% 33.5M/33.5M [00:00<00:00, 60.7MB/s]
!unzip withwithout-mask.zip
Archive: withwithout-mask.zip
inflating: maskdata/maskdata/test/with_mask/0-with-mask.jpg
inflating: maskdata/maskdata/test/with_mask/1-with-mask.jpg
...
inflating: masks2.0/masks/train/1/augmented_image_257.jpg
inflating: masks2.0/masks/train/1/augmented_image_261.jpg
inflating: masks2.0/masks/train/1/augmented_image_265.jpg
inflating: masks2.0/masks/train/1/augmented_image_270.jpg
inflating: masks2.0/masks/train/1/augmented_image_281.jpg
inflating: masks2.0/masks/train/1/augmented_image_286.jpg
inflating: masks2.0/masks/train/1/augmented_image_291.jpg
inflating: masks2.0/masks/train/1/augmented_image_296.jpg
inflating: masks2.0/masks/train/1/augmented_image_305.jpg
inflating: masks2.0/masks/train/1/augmented_image_306.jpg
inflating: masks2.0/masks/train/1/augmented_image_314.jpg
inflating: masks2.0/masks/train/1/pra2.jpg
!mkdir with-and-without-mask
!mv mask* with-and-without-mask
!ls
annotations kaggle.json
face-mask-12k-images-dataset.zip sample_data
'Face Mask Dataset' with-and-without-mask
face-mask-detection.zip withwithout-mask.zip
images
!kaggle datasets download -d lalitharajesh/haarcascades
Downloading haarcascades.zip to /content
0% 0.00/1.45M [00:00<?, ?B/s]
100% 1.45M/1.45M [00:00<00:00, 48.8MB/s]
!unzip haarcascades.zip
Archive: haarcascades.zip
inflating: haarcascade_eye.xml
inflating: haarcascade_eye_tree_eyeglasses.xml
inflating: haarcascade_frontalcatface.xml
inflating: haarcascade_frontalcatface_extended.xml
inflating: haarcascade_frontalface_alt.xml
inflating: haarcascade_frontalface_alt2.xml
inflating: haarcascade_frontalface_alt_tree.xml
inflating: haarcascade_frontalface_default.xml
inflating: haarcascade_fullbody.xml
inflating: haarcascade_lefteye_2splits.xml
inflating: haarcascade_licence_plate_rus_16stages.xml
inflating: haarcascade_lowerbody.xml
inflating: haarcascade_profileface.xml
inflating: haarcascade_righteye_2splits.xml
inflating: haarcascade_russian_plate_number.xml
inflating: haarcascade_smile.xml
inflating: haarcascade_upperbody.xml
!mkdir frontalface
!mv haarcascade* frontalface
!ls
annotations frontalface with-and-without-mask
face-mask-12k-images-dataset.zip images withwithout-mask.zip
'Face Mask Dataset' kaggle.json
face-mask-detection.zip sample_data
!cd with-and-without-mask
!ls
annotations frontalface with-and-without-mask
face-mask-12k-images-dataset.zip images withwithout-mask.zip
'Face Mask Dataset' kaggle.json
face-mask-detection.zip sample_data
!kaggle datasets download -d omkargurav/face-mask-dataset
!unzip face-mask-dataset.zip
!ls
[1;30;43m스트리밍 출력 내용이 길어서 마지막 5000줄이 삭제되었습니다.[0m
inflating: data/with_mask/with_mask_3301.jpg
inflating: data/with_mask/with_mask_3302.jpg
inflating: data/with_mask/with_mask_3303.jpg
inflating: data/with_mask/with_mask_3304.jpg
....
inflating: data/without_mask/without_mask_997.jpg
inflating: data/without_mask/without_mask_998.jpg
inflating: data/without_mask/without_mask_999.jpg
annotations frontalface
data images
face-mask-12k-images-dataset.zip kaggle.json
'Face Mask Dataset' sample_data
face-mask-dataset.zip with-and-without-mask
face-mask-detection.zip withwithout-mask.zip
import numpy as np # linear algebra
import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)
import matplotlib.pyplot as plt
import seaborn as sns
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.preprocessing.image import load_img
import cv2
import random
import glob
import torch
import shutil
import itertools
import numpy as np # linear algebra
import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)
import seaborn as sns
import matplotlib.pyplot as plt
from torch import nn
from torch import optim
from torchvision import transforms, datasets, models
from keras.models import Sequential
from keras.applications.vgg19 import preprocess_input
from keras.preprocessing.image import ImageDataGenerator
from keras.layers import Flatten, Dense, Conv2D, BatchNormalization, MaxPooling2D, Dropout
from tensorflow.keras.applications import EfficientNetB1, VGG19, ResNet50, InceptionV3, MobileNet, DenseNet201, NASNetMobile
import os
os.listdir("Face Mask Dataset/Train/")
['WithoutMask', 'WithMask']
print("Images in Train Dataset:\n")
print("Number of Images for with Mask category:{}".format(len(os.listdir("Face Mask Dataset/Train/WithMask"))))
print("Number of Images for with WithoutMask category:{}".format(len(os.listdir("Face Mask Dataset/Train/WithoutMask/"))))
Images in Train Dataset:
Number of Images for with Mask category:5000
Number of Images for with WithoutMask category:5000
train_dir = "Face Mask Dataset/Train/"
test_dir = "Face Mask Dataset/Test/"
Data Augmentation
Images:
#with Mask
plt.figure(figsize=(12,7))
for i in range(5):
sample = random.choice(os.listdir(train_dir+"WithMask/"))
plt.subplot(1,5,i+1)
img = load_img(train_dir+"WithMask/"+sample)
plt.subplots_adjust(hspace=0.001)
plt.xlabel("With Mask")
plt.imshow(img)
plt.show()
#without Mask
plt.figure(figsize=(12,7))
for i in range(5):
sample = random.choice(os.listdir(train_dir+"WithoutMask/"))
plt.subplot(1,5,i+1)
img = load_img(train_dir+"WithoutMask/"+sample)
plt.subplots_adjust(hspace=0.001)
plt.xlabel("Without Mask")
plt.imshow(img)
plt.show()
Data Augmentation:
height = 150
width=150
train_datagen = ImageDataGenerator(rescale=1.0/255,validation_split=0.2,shear_range = 0.2,zoom_range=0.2,horizontal_flip=True)
train = train_datagen.flow_from_directory(directory=train_dir,target_size=(height,width),
class_mode="categorical",batch_size=32,subset = "training")
valid_datagen = ImageDataGenerator(rescale=1.0/255)
valid = train_datagen.flow_from_directory(directory=train_dir,target_size=(height,width),
class_mode="categorical",batch_size=32,subset="validation")
Found 8000 images belonging to 2 classes.
Found 2000 images belonging to 2 classes.
Train 7 Models
ConvNet Testing layers with ConvNet
from keras.models import Sequential
histories = []
for i in range(3):
model = Sequential()
model.add(Conv2D(32, kernel_size=(3, 3), activation='relu', input_shape=(150, 150, 3)))
model.add(BatchNormalization())
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.2))
if i > 0:
model.add(Conv2D(64, kernel_size=(3, 3), activation='relu', input_shape=(150, 150, 3)))
model.add(BatchNormalization())
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.2))
if i > 1:
model.add(Conv2D(128, kernel_size=(3, 3), activation='relu', input_shape=(150, 150, 3)))
model.add(BatchNormalization())
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.2))
model.add(Flatten())
model.add(Dense(2, activation='sigmoid'))
model.summary()
model.compile(optimizer='adam', loss='binary_crossentropy', metrics='accuracy')
histories.append(model.fit_generator(generator=train,
validation_data=valid,
steps_per_epoch=len(train)//3,
validation_steps=len(valid)//3,
epochs=10))
Model: "sequential_17"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d_25 (Conv2D) (None, 148, 148, 32) 896
_________________________________________________________________
batch_normalization_26 (Batc (None, 148, 148, 32) 128
_________________________________________________________________
max_pooling2d_26 (MaxPooling (None, 74, 74, 32) 0
_________________________________________________________________
dropout_26 (Dropout) (None, 74, 74, 32) 0
_________________________________________________________________
flatten_13 (Flatten) (None, 175232) 0
_________________________________________________________________
dense_13 (Dense) (None, 2) 350466
=================================================================
Total params: 351,490
Trainable params: 351,426
Non-trainable params: 64
_________________________________________________________________
Epoch 1/10
/usr/local/lib/python3.7/dist-packages/keras/engine/training.py:1915: UserWarning: `Model.fit_generator` is deprecated and will be removed in a future version. Please use `Model.fit`, which supports generators.
warnings.warn('`Model.fit_generator` is deprecated and '
83/83 [==============================] - 19s 219ms/step - loss: 1.2393 - accuracy: 0.8840 - val_loss: 0.5732 - val_accuracy: 0.7783
Epoch 2/10
83/83 [==============================] - 18s 214ms/step - loss: 0.3196 - accuracy: 0.9602 - val_loss: 0.9409 - val_accuracy: 0.6577
Epoch 3/10
83/83 [==============================] - 18s 215ms/step - loss: 0.2628 - accuracy: 0.9671 - val_loss: 0.2437 - val_accuracy: 0.9241
Epoch 4/10
83/83 [==============================] - 18s 213ms/step - loss: 0.2200 - accuracy: 0.9718 - val_loss: 0.2475 - val_accuracy: 0.9077
Epoch 5/10
83/83 [==============================] - 18s 213ms/step - loss: 0.1381 - accuracy: 0.9792 - val_loss: 0.2589 - val_accuracy: 0.9301
Epoch 6/10
83/83 [==============================] - 18s 215ms/step - loss: 0.1896 - accuracy: 0.9735 - val_loss: 0.8904 - val_accuracy: 0.8824
Epoch 7/10
83/83 [==============================] - 18s 214ms/step - loss: 0.2077 - accuracy: 0.9714 - val_loss: 0.4959 - val_accuracy: 0.9509
Epoch 8/10
83/83 [==============================] - 18s 212ms/step - loss: 0.2023 - accuracy: 0.9685 - val_loss: 0.3304 - val_accuracy: 0.9494
Epoch 9/10
83/83 [==============================] - 18s 215ms/step - loss: 0.1309 - accuracy: 0.9788 - val_loss: 0.6745 - val_accuracy: 0.9137
Epoch 10/10
83/83 [==============================] - 18s 212ms/step - loss: 0.2494 - accuracy: 0.9642 - val_loss: 0.6319 - val_accuracy: 0.9286
Model: "sequential_18"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d_26 (Conv2D) (None, 148, 148, 32) 896
_________________________________________________________________
batch_normalization_27 (Batc (None, 148, 148, 32) 128
_________________________________________________________________
max_pooling2d_27 (MaxPooling (None, 74, 74, 32) 0
_________________________________________________________________
dropout_27 (Dropout) (None, 74, 74, 32) 0
_________________________________________________________________
conv2d_27 (Conv2D) (None, 72, 72, 64) 18496
_________________________________________________________________
batch_normalization_28 (Batc (None, 72, 72, 64) 256
_________________________________________________________________
max_pooling2d_28 (MaxPooling (None, 36, 36, 64) 0
_________________________________________________________________
dropout_28 (Dropout) (None, 36, 36, 64) 0
_________________________________________________________________
flatten_14 (Flatten) (None, 82944) 0
_________________________________________________________________
dense_14 (Dense) (None, 2) 165890
=================================================================
Total params: 185,666
Trainable params: 185,474
Non-trainable params: 192
_________________________________________________________________
Epoch 1/10
83/83 [==============================] - 20s 227ms/step - loss: 1.4548 - accuracy: 0.8697 - val_loss: 3.6322 - val_accuracy: 0.5268
Epoch 2/10
83/83 [==============================] - 18s 221ms/step - loss: 0.4808 - accuracy: 0.9614 - val_loss: 3.1423 - val_accuracy: 0.5253
Epoch 3/10
83/83 [==============================] - 18s 219ms/step - loss: 0.7233 - accuracy: 0.9539 - val_loss: 8.4443 - val_accuracy: 0.5372
Epoch 4/10
83/83 [==============================] - 18s 220ms/step - loss: 0.4472 - accuracy: 0.9629 - val_loss: 5.4666 - val_accuracy: 0.6503
Epoch 5/10
83/83 [==============================] - 18s 222ms/step - loss: 0.5129 - accuracy: 0.9696 - val_loss: 0.6152 - val_accuracy: 0.9464
Epoch 6/10
83/83 [==============================] - 18s 221ms/step - loss: 0.4910 - accuracy: 0.9682 - val_loss: 0.6641 - val_accuracy: 0.9479
Epoch 7/10
83/83 [==============================] - 18s 220ms/step - loss: 0.3646 - accuracy: 0.9815 - val_loss: 1.7643 - val_accuracy: 0.9018
Epoch 8/10
83/83 [==============================] - 18s 221ms/step - loss: 0.4164 - accuracy: 0.9763 - val_loss: 1.1468 - val_accuracy: 0.9211
Epoch 9/10
83/83 [==============================] - 18s 220ms/step - loss: 0.4656 - accuracy: 0.9766 - val_loss: 1.0286 - val_accuracy: 0.9494
Epoch 10/10
83/83 [==============================] - 18s 220ms/step - loss: 0.2667 - accuracy: 0.9871 - val_loss: 1.3896 - val_accuracy: 0.9494
Model: "sequential_19"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d_28 (Conv2D) (None, 148, 148, 32) 896
_________________________________________________________________
batch_normalization_29 (Batc (None, 148, 148, 32) 128
_________________________________________________________________
max_pooling2d_29 (MaxPooling (None, 74, 74, 32) 0
_________________________________________________________________
dropout_29 (Dropout) (None, 74, 74, 32) 0
_________________________________________________________________
conv2d_29 (Conv2D) (None, 72, 72, 64) 18496
_________________________________________________________________
batch_normalization_30 (Batc (None, 72, 72, 64) 256
_________________________________________________________________
max_pooling2d_30 (MaxPooling (None, 36, 36, 64) 0
_________________________________________________________________
dropout_30 (Dropout) (None, 36, 36, 64) 0
_________________________________________________________________
conv2d_30 (Conv2D) (None, 34, 34, 128) 73856
_________________________________________________________________
batch_normalization_31 (Batc (None, 34, 34, 128) 512
_________________________________________________________________
max_pooling2d_31 (MaxPooling (None, 17, 17, 128) 0
_________________________________________________________________
dropout_31 (Dropout) (None, 17, 17, 128) 0
_________________________________________________________________
flatten_15 (Flatten) (None, 36992) 0
_________________________________________________________________
dense_15 (Dense) (None, 2) 73986
=================================================================
Total params: 168,130
Trainable params: 167,682
Non-trainable params: 448
_________________________________________________________________
Epoch 1/10
83/83 [==============================] - 20s 229ms/step - loss: 0.5875 - accuracy: 0.9091 - val_loss: 3.0262 - val_accuracy: 0.4762
Epoch 2/10
83/83 [==============================] - 18s 221ms/step - loss: 0.2056 - accuracy: 0.9687 - val_loss: 0.4334 - val_accuracy: 0.8259
Epoch 3/10
83/83 [==============================] - 18s 222ms/step - loss: 0.1600 - accuracy: 0.9792 - val_loss: 0.7784 - val_accuracy: 0.8408
Epoch 4/10
83/83 [==============================] - 18s 223ms/step - loss: 0.1560 - accuracy: 0.9798 - val_loss: 0.4578 - val_accuracy: 0.9405
Epoch 5/10
83/83 [==============================] - 18s 221ms/step - loss: 0.2043 - accuracy: 0.9751 - val_loss: 1.5926 - val_accuracy: 0.8110
Epoch 6/10
83/83 [==============================] - 18s 221ms/step - loss: 0.2302 - accuracy: 0.9763 - val_loss: 0.5961 - val_accuracy: 0.9390
Epoch 7/10
83/83 [==============================] - 18s 220ms/step - loss: 0.0967 - accuracy: 0.9854 - val_loss: 1.3696 - val_accuracy: 0.8661
Epoch 8/10
83/83 [==============================] - 18s 221ms/step - loss: 0.1184 - accuracy: 0.9851 - val_loss: 0.5381 - val_accuracy: 0.9435
Epoch 9/10
83/83 [==============================] - 18s 219ms/step - loss: 0.1724 - accuracy: 0.9810 - val_loss: 15.4781 - val_accuracy: 0.5149
Epoch 10/10
83/83 [==============================] - 18s 220ms/step - loss: 0.1163 - accuracy: 0.9894 - val_loss: 0.2193 - val_accuracy: 0.9762
fig, axes = plt.subplots(2, 2, figsize=(15, 10))
for metric in histories[0].history:
index = list(histories[0].history).index(metric)
ax = axes.flatten()[index]
layer_num = 0
for history in histories:
layer_num += 1
ax.plot(history.history[metric], label=str(layer_num)+' layer(s)')
ax.set_title(metric)
ax.legend()
plt.show()
ConvNet
model_histories = []
for layer in [Conv2D(32, kernel_size=(3, 3), activation='relu', input_shape=(150, 150, 3))]:
model = Sequential()
model.add(layer)
model.add(BatchNormalization())
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.2))
model.add(Flatten())
model.add(Dense(2, activation='sigmoid'))
model.summary()
model.compile(optimizer='adam', loss='binary_crossentropy', metrics='accuracy')
model_histories.append(model.fit_generator(generator=train,
validation_data=valid,
steps_per_epoch=len(train)//3,
validation_steps=len(valid)//3,
epochs=10))
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d (Conv2D) (None, 148, 148, 32) 896
_________________________________________________________________
batch_normalization (BatchNo (None, 148, 148, 32) 128
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 74, 74, 32) 0
_________________________________________________________________
dropout (Dropout) (None, 74, 74, 32) 0
_________________________________________________________________
flatten (Flatten) (None, 175232) 0
_________________________________________________________________
dense (Dense) (None, 2) 350466
=================================================================
Total params: 351,490
Trainable params: 351,426
Non-trainable params: 64
_________________________________________________________________
/usr/local/lib/python3.7/dist-packages/keras/engine/training.py:1915: UserWarning: `Model.fit_generator` is deprecated and will be removed in a future version. Please use `Model.fit`, which supports generators.
warnings.warn('`Model.fit_generator` is deprecated and '
Epoch 1/10
83/83 [==============================] - 62s 214ms/step - loss: 0.9515 - accuracy: 0.8875 - val_loss: 0.3282 - val_accuracy: 0.8943
Epoch 2/10
83/83 [==============================] - 17s 209ms/step - loss: 0.2300 - accuracy: 0.9663 - val_loss: 0.4833 - val_accuracy: 0.7842
Epoch 3/10
83/83 [==============================] - 17s 208ms/step - loss: 0.2003 - accuracy: 0.9739 - val_loss: 0.3448 - val_accuracy: 0.9182
Epoch 4/10
83/83 [==============================] - 17s 211ms/step - loss: 0.2527 - accuracy: 0.9627 - val_loss: 0.3171 - val_accuracy: 0.9077
Epoch 5/10
83/83 [==============================] - 18s 223ms/step - loss: 0.3326 - accuracy: 0.9537 - val_loss: 0.7074 - val_accuracy: 0.8527
Epoch 6/10
83/83 [==============================] - 18s 222ms/step - loss: 0.2556 - accuracy: 0.9740 - val_loss: 0.2292 - val_accuracy: 0.9554
Epoch 7/10
83/83 [==============================] - 18s 220ms/step - loss: 0.2083 - accuracy: 0.9746 - val_loss: 0.3235 - val_accuracy: 0.9554
Epoch 8/10
83/83 [==============================] - 18s 212ms/step - loss: 0.1827 - accuracy: 0.9766 - val_loss: 0.2851 - val_accuracy: 0.9449
Epoch 9/10
83/83 [==============================] - 19s 224ms/step - loss: 0.1357 - accuracy: 0.9767 - val_loss: 0.3161 - val_accuracy: 0.9524
Epoch 10/10
83/83 [==============================] - 18s 221ms/step - loss: 0.1169 - accuracy: 0.9829 - val_loss: 0.3526 - val_accuracy: 0.9554
model.save('convnet.h5')
convnet_model = model
convnet_evaluate = convnet_model.evaluate_generator(valid)
convnet_evaluate
/usr/local/lib/python3.7/dist-packages/keras/engine/training.py:1948: UserWarning: `Model.evaluate_generator` is deprecated and will be removed in a future version. Please use `Model.evaluate`, which supports generators.
warnings.warn('`Model.evaluate_generator` is deprecated and '
[0.32139459252357483, 0.9520000219345093]
from IPython.display import SVG
from keras.utils.vis_utils import model_to_dot
SVG(model_to_dot(model, show_shapes=True, dpi=65).create(prog='dot', format='svg'))
MobileNet
Model Building:
from tensorflow.keras.applications.mobilenet_v2 import MobileNetV2
mobilenet = MobileNetV2(weights = "imagenet",include_top = False,input_shape=(150,150,3))
WARNING:tensorflow:`input_shape` is undefined or non-square, or `rows` is not in [96, 128, 160, 192, 224]. Weights for input shape (224, 224) will be loaded as the default.
Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/mobilenet_v2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.0_224_no_top.h5
9412608/9406464 [==============================] - 0s 0us/step
for layer in mobilenet.layers:
layer.trainable = False
from tensorflow.keras import Sequential
from tensorflow.keras.layers import Flatten,Dense
model = Sequential()
model.add(mobilenet)
model.add(Flatten())
model.add(Dense(2,activation="sigmoid"))
model.summary()
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
mobilenetv2_1.00_224 (Functi (None, 5, 5, 1280) 2257984
_________________________________________________________________
flatten (Flatten) (None, 32000) 0
_________________________________________________________________
dense (Dense) (None, 2) 64002
=================================================================
Total params: 2,321,986
Trainable params: 64,002
Non-trainable params: 2,257,984
_________________________________________________________________
model.compile(optimizer="adam",loss="binary_crossentropy",metrics ="accuracy")
# from tensorflow.keras.callbacks import ModelCheckpoint,EarlyStopping
# checkpoint = ModelCheckpoint("moblenet_facemask.h5",monitor="val_accuracy",save_best_only=True,verbose=1)
# earlystop = EarlyStopping(monitor="val_acc",patience=5,verbose=1)
model_histories.append(model.fit_generator(generator=train,steps_per_epoch=len(train)// 32,validation_data=valid,
validation_steps = len(valid)//32,epochs=10))
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py:1940: UserWarning: `Model.fit_generator` is deprecated and will be removed in a future version. Please use `Model.fit`, which supports generators.
warnings.warn('`Model.fit_generator` is deprecated and '
Epoch 1/10
7/7 [==============================] - 5s 366ms/step - loss: 1.3101 - accuracy: 0.7723 - val_loss: 0.8136 - val_accuracy: 0.8125
Epoch 2/10
7/7 [==============================] - 2s 259ms/step - loss: 0.5490 - accuracy: 0.9241 - val_loss: 0.0075 - val_accuracy: 1.0000
Epoch 3/10
7/7 [==============================] - 2s 245ms/step - loss: 0.1397 - accuracy: 0.9866 - val_loss: 0.0060 - val_accuracy: 1.0000
Epoch 4/10
7/7 [==============================] - 2s 256ms/step - loss: 0.0944 - accuracy: 0.9866 - val_loss: 0.0113 - val_accuracy: 1.0000
Epoch 5/10
7/7 [==============================] - 2s 255ms/step - loss: 0.1862 - accuracy: 0.9821 - val_loss: 0.2993 - val_accuracy: 0.9688
Epoch 6/10
7/7 [==============================] - 2s 240ms/step - loss: 0.0207 - accuracy: 0.9911 - val_loss: 0.0068 - val_accuracy: 1.0000
Epoch 7/10
7/7 [==============================] - 2s 257ms/step - loss: 0.0604 - accuracy: 0.9821 - val_loss: 5.8131e-05 - val_accuracy: 1.0000
Epoch 8/10
7/7 [==============================] - 2s 255ms/step - loss: 0.1621 - accuracy: 0.9866 - val_loss: 0.0020 - val_accuracy: 1.0000
Epoch 9/10
7/7 [==============================] - 2s 251ms/step - loss: 0.0516 - accuracy: 0.9911 - val_loss: 0.0098 - val_accuracy: 1.0000
Epoch 10/10
7/7 [==============================] - 2s 253ms/step - loss: 0.0249 - accuracy: 0.9955 - val_loss: 0.1287 - val_accuracy: 0.9688
mobile_evaluate = model.evaluate_generator(valid)
mobile_evaluate
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py:1973: UserWarning: `Model.evaluate_generator` is deprecated and will be removed in a future version. Please use `Model.evaluate`, which supports generators.
warnings.warn('`Model.evaluate_generator` is deprecated and '
[0.05435192584991455, 0.9879999756813049]
model.save("mobilenet.h5")
# pred = model.predict_classes(valid)
# pred[:15]
mobilenet_model = model
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/utils/generic_utils.py:497: CustomMaskWarning: Custom mask layers require a config and must override get_config. When loading, the custom mask layer must be passed to the custom_objects argument.
category=CustomMaskWarning)
from IPython.display import SVG
from tensorflow.python.keras.utils.vis_utils import model_to_dot
SVG(model_to_dot(mobilenet_model, show_shapes=True, dpi=65).create(prog='dot', format='svg'))
DenseNet
Model Building:
from tensorflow.keras.applications import DenseNet201
densenet = DenseNet201(weights = "imagenet",include_top = False,input_shape=(150,150,3))
Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/densenet/densenet201_weights_tf_dim_ordering_tf_kernels_notop.h5
74842112/74836368 [==============================] - 1s 0us/step
# freeze
for layer in densenet.layers:
layer.trainable = False
from tensorflow.keras import Sequential
from tensorflow.keras.layers import Flatten,Dense
model = Sequential()
model.add(densenet)
model.add(Flatten())
model.add(Dense(2,activation="sigmoid"))
model.summary()
Model: "sequential_1"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
densenet201 (Functional) (None, 4, 4, 1920) 18321984
_________________________________________________________________
flatten_1 (Flatten) (None, 30720) 0
_________________________________________________________________
dense_1 (Dense) (None, 2) 61442
=================================================================
Total params: 18,383,426
Trainable params: 61,442
Non-trainable params: 18,321,984
_________________________________________________________________
model.compile(optimizer="adam",loss="binary_crossentropy",metrics ="accuracy")
# from tensorflow.keras.callbacks import ModelCheckpoint,EarlyStopping
# checkpoint = ModelCheckpoint("densenet_facemask.h5",monitor="val_accuracy",save_best_only=True,verbose=1)
# earlystop = EarlyStopping(monitor="val_accuracy",patience=5,verbose=1)
model_histories.append(model.fit_generator(generator=train,steps_per_epoch=len(train)// 32,validation_data=valid,
validation_steps = len(valid)//32, epochs=10))
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py:1940: UserWarning: `Model.fit_generator` is deprecated and will be removed in a future version. Please use `Model.fit`, which supports generators.
warnings.warn('`Model.fit_generator` is deprecated and '
Epoch 1/10
7/7 [==============================] - 15s 677ms/step - loss: 0.3663 - accuracy: 0.8661 - val_loss: 0.0110 - val_accuracy: 1.0000
Epoch 2/10
7/7 [==============================] - 2s 264ms/step - loss: 0.0356 - accuracy: 0.9866 - val_loss: 5.8067e-05 - val_accuracy: 1.0000
Epoch 3/10
7/7 [==============================] - 2s 261ms/step - loss: 0.0371 - accuracy: 0.9955 - val_loss: 0.0018 - val_accuracy: 1.0000
Epoch 4/10
7/7 [==============================] - 2s 263ms/step - loss: 0.1412 - accuracy: 0.9821 - val_loss: 0.2351 - val_accuracy: 0.9688
Epoch 5/10
7/7 [==============================] - 2s 275ms/step - loss: 0.0040 - accuracy: 1.0000 - val_loss: 4.4025e-04 - val_accuracy: 1.0000
Epoch 6/10
7/7 [==============================] - 2s 272ms/step - loss: 0.0485 - accuracy: 0.9955 - val_loss: 5.6109e-06 - val_accuracy: 1.0000
Epoch 7/10
7/7 [==============================] - 2s 277ms/step - loss: 0.0391 - accuracy: 0.9777 - val_loss: 0.0014 - val_accuracy: 1.0000
Epoch 8/10
7/7 [==============================] - 2s 262ms/step - loss: 0.0904 - accuracy: 0.9821 - val_loss: 0.0164 - val_accuracy: 1.0000
Epoch 9/10
7/7 [==============================] - 2s 266ms/step - loss: 0.0157 - accuracy: 0.9955 - val_loss: 0.0150 - val_accuracy: 1.0000
Epoch 10/10
7/7 [==============================] - 2s 272ms/step - loss: 0.0234 - accuracy: 0.9955 - val_loss: 0.0038 - val_accuracy: 1.0000
dense_evaluate = model.evaluate_generator(valid)
dense_evaluate
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py:1973: UserWarning: `Model.evaluate_generator` is deprecated and will be removed in a future version. Please use `Model.evaluate`, which supports generators.
warnings.warn('`Model.evaluate_generator` is deprecated and '
[0.03043755330145359, 0.9919999837875366]
model.save("densenet.h5")
densenet_model = model
#pred = model.predict_classes(valid)
#pred[:15]
SVG(model_to_dot(densenet_model, show_shapes=True, dpi=65).create(prog='dot', format='svg'))
InceptionV3
Model Building:
from tensorflow.keras.applications import InceptionV3
inception = InceptionV3(include_top=False, input_shape=(150, 150, 3))
Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/inception_v3/inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5
87916544/87910968 [==============================] - 0s 0us/step
# freeze
for layer in inception.layers:
layer.trainable = False
from tensorflow.keras import Sequential
from tensorflow.keras.layers import Flatten,Dense
model = Sequential()
model.add(inception)
model.add(Flatten())
model.add(Dense(2,activation="sigmoid"))
model.summary()
Model: "sequential_2"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
inception_v3 (Functional) (None, 3, 3, 2048) 21802784
_________________________________________________________________
flatten_2 (Flatten) (None, 18432) 0
_________________________________________________________________
dense_2 (Dense) (None, 2) 36866
=================================================================
Total params: 21,839,650
Trainable params: 36,866
Non-trainable params: 21,802,784
_________________________________________________________________
model.compile(optimizer="adam",loss="binary_crossentropy",metrics ="accuracy")
# from tensorflow.keras.callbacks import ModelCheckpoint,EarlyStopping
# checkpoint = ModelCheckpoint("densenet_facemask.h5",monitor="val_accuracy",save_best_only=True,verbose=1)
# earlystop = EarlyStopping(monitor="val_accuracy",patience=5,verbose=1)
model_histories.append(model.fit_generator(generator=train,steps_per_epoch=len(train)// 32,validation_data=valid,
validation_steps = len(valid)//32, epochs=10))
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py:1940: UserWarning: `Model.fit_generator` is deprecated and will be removed in a future version. Please use `Model.fit`, which supports generators.
warnings.warn('`Model.fit_generator` is deprecated and '
Epoch 1/10
7/7 [==============================] - 7s 429ms/step - loss: 0.3650 - accuracy: 0.8616 - val_loss: 0.1306 - val_accuracy: 0.9375
Epoch 2/10
7/7 [==============================] - 2s 266ms/step - loss: 0.0570 - accuracy: 0.9866 - val_loss: 0.4049 - val_accuracy: 0.9688
Epoch 3/10
7/7 [==============================] - 2s 252ms/step - loss: 0.1229 - accuracy: 0.9866 - val_loss: 2.2869e-06 - val_accuracy: 1.0000
Epoch 4/10
7/7 [==============================] - 2s 251ms/step - loss: 0.0541 - accuracy: 0.9911 - val_loss: 8.2029e-04 - val_accuracy: 1.0000
Epoch 5/10
7/7 [==============================] - 2s 256ms/step - loss: 0.0031 - accuracy: 1.0000 - val_loss: 1.2229e-04 - val_accuracy: 1.0000
Epoch 6/10
7/7 [==============================] - 2s 250ms/step - loss: 0.0035 - accuracy: 1.0000 - val_loss: 3.8108e-04 - val_accuracy: 1.0000
Epoch 7/10
7/7 [==============================] - 2s 250ms/step - loss: 0.3397 - accuracy: 0.9732 - val_loss: 0.0016 - val_accuracy: 1.0000
Epoch 8/10
7/7 [==============================] - 2s 262ms/step - loss: 0.0755 - accuracy: 0.9955 - val_loss: 0.0206 - val_accuracy: 1.0000
Epoch 9/10
7/7 [==============================] - 2s 258ms/step - loss: 0.1150 - accuracy: 0.9911 - val_loss: 6.4188e-06 - val_accuracy: 1.0000
Epoch 10/10
7/7 [==============================] - 2s 249ms/step - loss: 0.1467 - accuracy: 0.9821 - val_loss: 6.5210e-04 - val_accuracy: 1.0000
inception_evaluate = model.evaluate_generator(valid)
inception_evaluate
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py:1973: UserWarning: `Model.evaluate_generator` is deprecated and will be removed in a future version. Please use `Model.evaluate`, which supports generators.
warnings.warn('`Model.evaluate_generator` is deprecated and '
[0.13425445556640625, 0.9775000214576721]
model.save("inception.h5")
inception_model = model
#pred = model.predict_classes(valid)
#pred[:15]
SVG(model_to_dot(inception_model, show_shapes=True, dpi=65).create(prog='dot', format='svg'))
VGG19
Model Building:
from tensorflow.keras.applications import InceptionV3
vgg19 = VGG19(include_top=False, input_shape=(150, 150, 3))
Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/vgg19/vgg19_weights_tf_dim_ordering_tf_kernels_notop.h5
80142336/80134624 [==============================] - 0s 0us/step
# freeze
for layer in vgg19.layers:
layer.trainable = False
from tensorflow.keras import Sequential
from tensorflow.keras.layers import Flatten,Dense
model = Sequential()
model.add(vgg19)
model.add(Flatten())
model.add(Dense(2,activation="sigmoid"))
model.summary()
Model: "sequential_3"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
vgg19 (Functional) (None, 4, 4, 512) 20024384
_________________________________________________________________
flatten_3 (Flatten) (None, 8192) 0
_________________________________________________________________
dense_3 (Dense) (None, 2) 16386
=================================================================
Total params: 20,040,770
Trainable params: 16,386
Non-trainable params: 20,024,384
_________________________________________________________________
model.compile(optimizer="adam",loss="binary_crossentropy",metrics ="accuracy")
# from tensorflow.keras.callbacks import ModelCheckpoint,EarlyStopping
# checkpoint = ModelCheckpoint("densenet_facemask.h5",monitor="val_accuracy",save_best_only=True,verbose=1)
# earlystop = EarlyStopping(monitor="val_accuracy",patience=5,verbose=1)
model_histories.append(model.fit_generator(generator=train,steps_per_epoch=len(train)// 32,validation_data=valid,
validation_steps = len(valid)//32, epochs=10))
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py:1940: UserWarning: `Model.fit_generator` is deprecated and will be removed in a future version. Please use `Model.fit`, which supports generators.
warnings.warn('`Model.fit_generator` is deprecated and '
Epoch 1/10
7/7 [==============================] - 6s 296ms/step - loss: 0.6241 - accuracy: 0.7143 - val_loss: 0.4291 - val_accuracy: 0.8438
Epoch 2/10
7/7 [==============================] - 2s 257ms/step - loss: 0.3762 - accuracy: 0.8616 - val_loss: 0.4027 - val_accuracy: 0.8438
Epoch 3/10
7/7 [==============================] - 2s 261ms/step - loss: 0.2947 - accuracy: 0.8929 - val_loss: 0.3127 - val_accuracy: 0.8750
Epoch 4/10
7/7 [==============================] - 2s 256ms/step - loss: 0.1942 - accuracy: 0.9464 - val_loss: 0.2254 - val_accuracy: 0.9375
Epoch 5/10
7/7 [==============================] - 2s 263ms/step - loss: 0.1968 - accuracy: 0.9330 - val_loss: 0.2045 - val_accuracy: 0.9375
Epoch 6/10
7/7 [==============================] - 2s 262ms/step - loss: 0.1478 - accuracy: 0.9643 - val_loss: 0.0764 - val_accuracy: 1.0000
Epoch 7/10
7/7 [==============================] - 2s 257ms/step - loss: 0.1300 - accuracy: 0.9777 - val_loss: 0.1285 - val_accuracy: 0.9688
Epoch 8/10
7/7 [==============================] - 2s 256ms/step - loss: 0.1249 - accuracy: 0.9732 - val_loss: 0.1094 - val_accuracy: 0.9688
Epoch 9/10
7/7 [==============================] - 2s 256ms/step - loss: 0.1324 - accuracy: 0.9509 - val_loss: 0.1518 - val_accuracy: 0.9375
Epoch 10/10
7/7 [==============================] - 2s 258ms/step - loss: 0.1596 - accuracy: 0.9420 - val_loss: 0.0807 - val_accuracy: 1.0000
vgg19_evaluate = model.evaluate_generator(valid)
vgg19_evaluate
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py:1973: UserWarning: `Model.evaluate_generator` is deprecated and will be removed in a future version. Please use `Model.evaluate`, which supports generators.
warnings.warn('`Model.evaluate_generator` is deprecated and '
[0.1290038824081421, 0.965499997138977]
model.save("vgg19.h5")
vgg19_model = model
#pred = model.predict_classes(valid)
#pred[:15]
SVG(model_to_dot(vgg19_model, show_shapes=True, dpi=65).create(prog='dot', format='svg'))
EfficientNet
Model Building:
from tensorflow.keras.applications import InceptionV3
efficientnet = EfficientNetB1(include_top=False, input_shape=(150, 150, 3))
Downloading data from https://storage.googleapis.com/keras-applications/efficientnetb1_notop.h5
27025408/27018416 [==============================] - 0s 0us/step
# freeze
for layer in efficientnet.layers:
layer.trainable = False
from tensorflow.keras import Sequential
from tensorflow.keras.layers import Flatten,Dense
model = Sequential()
model.add(efficientnet)
model.add(Flatten())
model.add(Dense(2,activation="sigmoid"))
model.summary()
Model: "sequential_4"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
efficientnetb1 (Functional) (None, 5, 5, 1280) 6575239
_________________________________________________________________
flatten_4 (Flatten) (None, 32000) 0
_________________________________________________________________
dense_4 (Dense) (None, 2) 64002
=================================================================
Total params: 6,639,241
Trainable params: 64,002
Non-trainable params: 6,575,239
_________________________________________________________________
model.compile(optimizer="adam",loss="binary_crossentropy",metrics ="accuracy")
# from tensorflow.keras.callbacks import ModelCheckpoint,EarlyStopping
# checkpoint = ModelCheckpoint("densenet_facemask.h5",monitor="val_accuracy",save_best_only=True,verbose=1)
# earlystop = EarlyStopping(monitor="val_accuracy",patience=5,verbose=1)
model_histories.append(model.fit_generator(generator=train,steps_per_epoch=len(train)// 32,validation_data=valid,
validation_steps = len(valid)//32, epochs=10))
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py:1940: UserWarning: `Model.fit_generator` is deprecated and will be removed in a future version. Please use `Model.fit`, which supports generators.
warnings.warn('`Model.fit_generator` is deprecated and '
Epoch 1/10
7/7 [==============================] - 10s 505ms/step - loss: 2.6089 - accuracy: 0.5045 - val_loss: 1.1192 - val_accuracy: 0.5000
Epoch 2/10
7/7 [==============================] - 2s 254ms/step - loss: 1.9411 - accuracy: 0.5000 - val_loss: 1.8807 - val_accuracy: 0.4375
Epoch 3/10
7/7 [==============================] - 2s 264ms/step - loss: 1.5564 - accuracy: 0.4955 - val_loss: 1.5464 - val_accuracy: 0.5625
Epoch 4/10
7/7 [==============================] - 2s 257ms/step - loss: 1.0811 - accuracy: 0.5179 - val_loss: 0.8355 - val_accuracy: 0.3750
Epoch 5/10
7/7 [==============================] - 2s 266ms/step - loss: 0.8195 - accuracy: 0.5580 - val_loss: 1.0967 - val_accuracy: 0.3750
Epoch 6/10
7/7 [==============================] - 2s 265ms/step - loss: 0.9739 - accuracy: 0.5223 - val_loss: 0.7223 - val_accuracy: 0.6250
Epoch 7/10
7/7 [==============================] - 2s 262ms/step - loss: 0.9429 - accuracy: 0.5179 - val_loss: 1.2246 - val_accuracy: 0.3125
Epoch 8/10
7/7 [==============================] - 2s 265ms/step - loss: 0.9037 - accuracy: 0.5045 - val_loss: 0.8160 - val_accuracy: 0.5000
Epoch 9/10
7/7 [==============================] - 2s 253ms/step - loss: 0.8912 - accuracy: 0.4196 - val_loss: 0.8048 - val_accuracy: 0.5312
Epoch 10/10
7/7 [==============================] - 2s 255ms/step - loss: 0.7982 - accuracy: 0.5089 - val_loss: 0.7111 - val_accuracy: 0.6250
efficientnet_evaluate = model.evaluate_generator(valid)
efficientnet_evaluate
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py:1973: UserWarning: `Model.evaluate_generator` is deprecated and will be removed in a future version. Please use `Model.evaluate`, which supports generators.
warnings.warn('`Model.evaluate_generator` is deprecated and '
[0.7567768692970276, 0.5]
model.save("efficientnet.h5")
efficientnet_model = model
#pred = model.predict_classes(valid)
#pred[:15]
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/utils/generic_utils.py:497: CustomMaskWarning: Custom mask layers require a config and must override get_config. When loading, the custom mask layer must be passed to the custom_objects argument.
category=CustomMaskWarning)
SVG(model_to_dot(efficientnet_model, show_shapes=True, dpi=65).create(prog='dot', format='svg'))
ResNet50
Model Building:
from tensorflow.keras.applications import ResNet50
resnet = ResNet50(include_top=False, input_shape=(150, 150, 3))
Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/resnet/resnet50_weights_tf_dim_ordering_tf_kernels_notop.h5
94773248/94765736 [==============================] - 1s 0us/step
# freeze
for layer in resnet.layers:
layer.trainable = False
from tensorflow.keras import Sequential
from tensorflow.keras.layers import Flatten,Dense
model = Sequential()
model.add(resnet)
model.add(Flatten())
model.add(Dense(2,activation="sigmoid"))
model.summary()
Model: "sequential_5"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
resnet50 (Functional) (None, 5, 5, 2048) 23587712
_________________________________________________________________
flatten_5 (Flatten) (None, 51200) 0
_________________________________________________________________
dense_5 (Dense) (None, 2) 102402
=================================================================
Total params: 23,690,114
Trainable params: 102,402
Non-trainable params: 23,587,712
_________________________________________________________________
model.compile(optimizer="adam",loss="binary_crossentropy",metrics ="accuracy")
# from tensorflow.keras.callbacks import ModelCheckpoint,EarlyStopping
# checkpoint = ModelCheckpoint("densenet_facemask.h5",monitor="val_accuracy",save_best_only=True,verbose=1)
# earlystop = EarlyStopping(monitor="val_accuracy",patience=5,verbose=1)
model_histories.append(model.fit_generator(generator=train,steps_per_epoch=len(train)// 32,validation_data=valid,
validation_steps = len(valid)//32, epochs=10))
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py:1940: UserWarning: `Model.fit_generator` is deprecated and will be removed in a future version. Please use `Model.fit`, which supports generators.
warnings.warn('`Model.fit_generator` is deprecated and '
Epoch 1/10
7/7 [==============================] - 6s 397ms/step - loss: 1.0013 - accuracy: 0.4732 - val_loss: 0.6948 - val_accuracy: 0.6250
Epoch 2/10
7/7 [==============================] - 2s 262ms/step - loss: 0.7961 - accuracy: 0.5536 - val_loss: 0.7714 - val_accuracy: 0.5000
Epoch 3/10
7/7 [==============================] - 2s 260ms/step - loss: 0.7350 - accuracy: 0.5759 - val_loss: 0.6794 - val_accuracy: 0.6875
Epoch 4/10
7/7 [==============================] - 2s 264ms/step - loss: 0.6719 - accuracy: 0.6071 - val_loss: 0.6982 - val_accuracy: 0.6250
Epoch 5/10
7/7 [==============================] - 2s 250ms/step - loss: 0.7277 - accuracy: 0.5089 - val_loss: 0.6723 - val_accuracy: 0.4688
Epoch 6/10
7/7 [==============================] - 2s 260ms/step - loss: 0.6965 - accuracy: 0.5625 - val_loss: 0.6064 - val_accuracy: 0.5938
Epoch 7/10
7/7 [==============================] - 2s 265ms/step - loss: 0.7266 - accuracy: 0.5045 - val_loss: 0.7173 - val_accuracy: 0.5312
Epoch 8/10
7/7 [==============================] - 2s 256ms/step - loss: 0.6169 - accuracy: 0.6920 - val_loss: 0.6500 - val_accuracy: 0.7812
Epoch 9/10
7/7 [==============================] - 2s 264ms/step - loss: 0.6377 - accuracy: 0.6250 - val_loss: 0.6285 - val_accuracy: 0.6562
Epoch 10/10
7/7 [==============================] - 2s 255ms/step - loss: 0.6223 - accuracy: 0.6741 - val_loss: 0.6416 - val_accuracy: 0.7188
resnet_evaluate = model.evaluate_generator(valid)
resnet_evaluate
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py:1973: UserWarning: `Model.evaluate_generator` is deprecated and will be removed in a future version. Please use `Model.evaluate`, which supports generators.
warnings.warn('`Model.evaluate_generator` is deprecated and '
[0.5898503065109253, 0.7705000042915344]
model.save("resnet.h5")
resnet_model = model
#pred = model.predict_classes(valid)
#pred[:15]
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/utils/generic_utils.py:497: CustomMaskWarning: Custom mask layers require a config and must override get_config. When loading, the custom mask layer must be passed to the custom_objects argument.
category=CustomMaskWarning)
SVG(model_to_dot(resnet_model, show_shapes=True, dpi=65).create(prog='dot', format='svg'))
ResNet50V2
Model Building:
from tensorflow.keras.applications import ResNet50V2
resnetv2 = ResNet50V2(include_top=False, input_shape=(150, 150, 3))
Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/resnet/resnet50v2_weights_tf_dim_ordering_tf_kernels_notop.h5
94674944/94668760 [==============================] - 1s 0us/step
# freeze
for layer in resnetv2.layers:
layer.trainable = False
from tensorflow.keras import Sequential
from tensorflow.keras.layers import Flatten,Dense
model = Sequential()
model.add(resnetv2)
model.add(Flatten())
model.add(Dense(2,activation="sigmoid"))
model.summary()
Model: "sequential_6"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
resnet50v2 (Functional) (None, 5, 5, 2048) 23564800
_________________________________________________________________
flatten_6 (Flatten) (None, 51200) 0
_________________________________________________________________
dense_6 (Dense) (None, 2) 102402
=================================================================
Total params: 23,667,202
Trainable params: 102,402
Non-trainable params: 23,564,800
_________________________________________________________________
model.compile(optimizer="adam",loss="binary_crossentropy",metrics ="accuracy")
# from tensorflow.keras.callbacks import ModelCheckpoint,EarlyStopping
# checkpoint = ModelCheckpoint("densenet_facemask.h5",monitor="val_accuracy",save_best_only=True,verbose=1)
# earlystop = EarlyStopping(monitor="val_accuracy",patience=5,verbose=1)
model_histories.append(model.fit_generator(generator=train,steps_per_epoch=len(train)// 32,validation_data=valid,
validation_steps = len(valid)//32, epochs=10))
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py:1940: UserWarning: `Model.fit_generator` is deprecated and will be removed in a future version. Please use `Model.fit`, which supports generators.
warnings.warn('`Model.fit_generator` is deprecated and '
Epoch 1/10
7/7 [==============================] - 5s 364ms/step - loss: 0.1961 - accuracy: 0.9196 - val_loss: 1.2123e-06 - val_accuracy: 1.0000
Epoch 2/10
7/7 [==============================] - 2s 251ms/step - loss: 0.0297 - accuracy: 0.9955 - val_loss: 0.5872 - val_accuracy: 0.9375
Epoch 3/10
7/7 [==============================] - 2s 255ms/step - loss: 0.1399 - accuracy: 0.9821 - val_loss: 0.3786 - val_accuracy: 0.9688
Epoch 4/10
7/7 [==============================] - 2s 254ms/step - loss: 0.1509 - accuracy: 0.9866 - val_loss: 3.3133e-07 - val_accuracy: 1.0000
Epoch 5/10
7/7 [==============================] - 2s 257ms/step - loss: 0.1544 - accuracy: 0.9911 - val_loss: 0.0814 - val_accuracy: 0.9688
Epoch 6/10
7/7 [==============================] - 2s 253ms/step - loss: 0.0488 - accuracy: 0.9955 - val_loss: 1.4253e-10 - val_accuracy: 1.0000
Epoch 7/10
7/7 [==============================] - 2s 254ms/step - loss: 0.0167 - accuracy: 0.9955 - val_loss: 4.7322e-06 - val_accuracy: 1.0000
Epoch 8/10
7/7 [==============================] - 2s 254ms/step - loss: 0.0468 - accuracy: 0.9955 - val_loss: 3.7470e-06 - val_accuracy: 1.0000
Epoch 9/10
7/7 [==============================] - 2s 250ms/step - loss: 0.0133 - accuracy: 0.9955 - val_loss: 0.0037 - val_accuracy: 1.0000
Epoch 10/10
7/7 [==============================] - 2s 256ms/step - loss: 0.0225 - accuracy: 0.9955 - val_loss: 2.0238e-10 - val_accuracy: 1.0000
resnetv2_evaluate = model.evaluate_generator(valid)
resnetv2_evaluate
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py:1973: UserWarning: `Model.evaluate_generator` is deprecated and will be removed in a future version. Please use `Model.evaluate`, which supports generators.
warnings.warn('`Model.evaluate_generator` is deprecated and '
[0.10062181204557419, 0.9890000224113464]
model.save("resnetv2.h5")
resnetv2_model = model
#pred = model.predict_classes(valid)
#pred[:15]
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/utils/generic_utils.py:497: CustomMaskWarning: Custom mask layers require a config and must override get_config. When loading, the custom mask layer must be passed to the custom_objects argument.
category=CustomMaskWarning)
SVG(model_to_dot(resnetv2_model, show_shapes=True, dpi=65).create(prog='dot', format='svg'))
Model Comparison
model_histories
[<keras.callbacks.History at 0x7f09426c7a50>,
<tensorflow.python.keras.callbacks.History at 0x7f09422356d0>,
<tensorflow.python.keras.callbacks.History at 0x7f0806099090>,
<tensorflow.python.keras.callbacks.History at 0x7f067efd9d10>,
<tensorflow.python.keras.callbacks.History at 0x7f067d0d6050>,
<tensorflow.python.keras.callbacks.History at 0x7f067b2fce50>,
<tensorflow.python.keras.callbacks.History at 0x7f06787aaed0>,
<tensorflow.python.keras.callbacks.History at 0x7f0676616b10>]
names=['ConvNet', 'MobileNet', 'DenseNet', 'InceptionV3', 'VGG19', 'EfficientNet', 'ResNet50', 'ResNet50V2']
fig, axes = plt.subplots(2, 2, figsize=(15, 10))
fig.subplots_adjust(hspace=0.3)
for metric in model_histories[0].history:
index = list(model_histories[0].history).index(metric)
ax = axes.flatten()[index]
name_index = 0
for history in model_histories:
ax.plot(history.history[metric], label=names[name_index])
name_index += 1
ax.set_title(metric+' over epochs', size=15)
ax.set_xlabel('epochs')
ax.set_ylabel(metric)
ax.legend()
plt.show()
df = pd. DataFrame([convnet_evaluate, mobile_evaluate, dense_evaluate, inception_evaluate, vgg19_evaluate, efficientnet_evaluate, resnet_evaluate, resnetv2_evaluate],
columns = ["loss", "accuracy"],
index = ['ConvNet', 'MobileNet', 'DenseNet', 'InceptionV3', 'VGG19', 'EfficientNet', 'ResNet50', 'ResNet50V2'])
df
| loss | accuracy | |
|---|---|---|
| ConvNet | 0.321395 | 0.9520 |
| MobileNet | 0.054352 | 0.9880 |
| DenseNet | 0.030438 | 0.9920 |
| InceptionV3 | 0.134254 | 0.9775 |
| VGG19 | 0.129004 | 0.9655 |
| EfficientNet | 0.756777 | 0.5000 |
| ResNet50 | 0.589850 | 0.7705 |
| ResNet50V2 | 0.100622 | 0.9890 |
## download trained models
from google.colab import files
files.download('convnet.h5') # from colab to browser download
files.download('mobilenet.h5')
files.download('densenet.h5')
files.download('inception.h5')
files.download('vgg19.h5')
files.download('efficientnet.h5')
files.download('resnet.h5')
files.download('resnetv2.h5')
<IPython.core.display.Javascript object>
<IPython.core.display.Javascript object>
<IPython.core.display.Javascript object>
<IPython.core.display.Javascript object>
<IPython.core.display.Javascript object>
<IPython.core.display.Javascript object>
<IPython.core.display.Javascript object>
<IPython.core.display.Javascript object>
<IPython.core.display.Javascript object>
<IPython.core.display.Javascript object>
<IPython.core.display.Javascript object>
<IPython.core.display.Javascript object>
<IPython.core.display.Javascript object>
<IPython.core.display.Javascript object>
<IPython.core.display.Javascript object>
<IPython.core.display.Javascript object>
Check Model Performance with OpenCV
# Load & Evaluate another face mask image
mask = "with-and-without-mask/"
plt.figure(figsize=(8,7))
label = {0:"With Mask",1:"Without Mask"}
color_label = {0: (0,255,0),1 : (0,0,255)}
cascade = cv2.CascadeClassifier("frontalface/haarcascade_frontalface_default.xml")
count = 0
i = "with-and-without-mask/maskdata/maskdata/test/without_mask/356.jpg"
frame =cv2.imread(i)
gray = cv2.cvtColor(frame,cv2.COLOR_BGR2GRAY)
faces = cascade.detectMultiScale(gray,1.1,4)
models = [convnet_model, inception_model, mobilenet_model, densenet_model, vgg19_model, efficientnet_model, resnetv2_model]
for model in models:
for x,y,w,h in faces:
face_image = frame[y:y+h,x:x+w]
resize_img = cv2.resize(face_image,(150,150))
normalized = resize_img/255.0
reshape = np.reshape(normalized,(1,150,150,3))
reshape = np.vstack([reshape])
if model == convnet_model:
name = 'ConvNet'
if model == inception_model:
name = 'InceptionV3'
if model == mobilenet_model:
name = 'MobileNet'
if model == densenet_model:
name = 'DenseNet'
if model == vgg19_model:
name = 'VGG19'
if model == efficientnet_model:
name = 'EfficientNet'
if model == resnetv2_model:
name = 'ResNet50V2'
result = model.predict_classes(reshape)
if result == 0:
cv2.rectangle(frame,(x,y),(x+w,y+h),color_label[0],3)
cv2.rectangle(frame,(x,y-50),(x+w,y),color_label[0],-1)
cv2.putText(frame,label[0],(x,y-10),cv2.FONT_HERSHEY_SIMPLEX,1,(255,255,255),2)
frame = cv2.cvtColor(frame,cv2.COLOR_BGR2RGB)
plt.imshow(frame)
elif result == 1:
cv2.rectangle(frame,(x,y),(x+w,y+h),color_label[1],3)
cv2.rectangle(frame,(x,y-50),(x+w,y),color_label[1],-1)
cv2.putText(frame,label[1],(x,y-10),cv2.FONT_HERSHEY_SIMPLEX,1,(255,255,255),2)
frame = cv2.cvtColor(frame,cv2.COLOR_BGR2RGB)
plt.imshow(frame)
plt.title(name)
plt.axis("off")
plt.show()
cv2.destroyAllWindows()
/usr/local/lib/python3.7/dist-packages/keras/engine/sequential.py:450: UserWarning: `model.predict_classes()` is deprecated and will be removed after 2021-01-01. Please use instead:* `np.argmax(model.predict(x), axis=-1)`, if your model does multi-class classification (e.g. if it uses a `softmax` last-layer activation).* `(model.predict(x) > 0.5).astype("int32")`, if your model does binary classification (e.g. if it uses a `sigmoid` last-layer activation).
warnings.warn('`model.predict_classes()` is deprecated and '
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/sequential.py:455: UserWarning: `model.predict_classes()` is deprecated and will be removed after 2021-01-01. Please use instead:* `np.argmax(model.predict(x), axis=-1)`, if your model does multi-class classification (e.g. if it uses a `softmax` last-layer activation).* `(model.predict(x) > 0.5).astype("int32")`, if your model does binary classification (e.g. if it uses a `sigmoid` last-layer activation).
warnings.warn('`model.predict_classes()` is deprecated and '
WARNING:tensorflow:5 out of the last 5 calls to <function Model.make_predict_function.<locals>.predict_function at 0x7f064e8d7170> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for more details.
WARNING:tensorflow:6 out of the last 6 calls to <function Model.make_predict_function.<locals>.predict_function at 0x7f064d37c440> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for more details.
#mask = "../input/with-and-without-mask/"
plt.figure(figsize=(8,7))
label = {0:"With Mask",1:"Without Mask"}
color_label = {0: (0,255,0),1 : (0,0,255)}
cascade = cv2.CascadeClassifier("frontalface/haarcascade_frontalface_default.xml")
count = 0
i = "with-and-without-mask/maskdata/maskdata/test/without_mask/431.jpg"
frame =cv2.imread(i)
gray = cv2.cvtColor(frame,cv2.COLOR_BGR2GRAY)
faces = cascade.detectMultiScale(gray,1.1,4)
models = [convnet_model, inception_model, mobilenet_model, densenet_model, vgg19_model, efficientnet_model, resnetv2_model]
for model in models:
for x,y,w,h in faces:
face_image = frame[y:y+h,x:x+w]
resize_img = cv2.resize(face_image,(150,150))
normalized = resize_img/255.0
reshape = np.reshape(normalized,(1,150,150,3))
reshape = np.vstack([reshape])
if model == convnet_model:
name = 'ConvNet'
if model == inception_model:
name = 'InceptionV3'
if model == mobilenet_model:
name = 'MobileNet'
if model == densenet_model:
name = 'DenseNet'
if model == vgg19_model:
name = 'VGG19'
if model == efficientnet_model:
name = 'EfficientNet'
if model == resnetv2_model:
name = 'ResNet50V2'
result = model.predict_classes(reshape)
if result == 0:
cv2.rectangle(frame,(x,y),(x+w,y+h),color_label[0],3)
cv2.rectangle(frame,(x,y-50),(x+w,y),color_label[0],-1)
cv2.putText(frame,label[0],(x,y-10),cv2.FONT_HERSHEY_SIMPLEX,1,(255,255,255),2)
frame = cv2.cvtColor(frame,cv2.COLOR_BGR2RGB)
plt.imshow(frame)
elif result == 1:
cv2.rectangle(frame,(x,y),(x+w,y+h),color_label[1],3)
cv2.rectangle(frame,(x,y-50),(x+w,y),color_label[1],-1)
cv2.putText(frame,label[1],(x,y-10),cv2.FONT_HERSHEY_SIMPLEX,1,(255,255,255),2)
frame = cv2.cvtColor(frame,cv2.COLOR_BGR2RGB)
plt.imshow(frame)
plt.title(name)
plt.axis("off")
plt.show()
cv2.destroyAllWindows()
/usr/local/lib/python3.7/dist-packages/keras/engine/sequential.py:450: UserWarning: `model.predict_classes()` is deprecated and will be removed after 2021-01-01. Please use instead:* `np.argmax(model.predict(x), axis=-1)`, if your model does multi-class classification (e.g. if it uses a `softmax` last-layer activation).* `(model.predict(x) > 0.5).astype("int32")`, if your model does binary classification (e.g. if it uses a `sigmoid` last-layer activation).
warnings.warn('`model.predict_classes()` is deprecated and '
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/sequential.py:455: UserWarning: `model.predict_classes()` is deprecated and will be removed after 2021-01-01. Please use instead:* `np.argmax(model.predict(x), axis=-1)`, if your model does multi-class classification (e.g. if it uses a `softmax` last-layer activation).* `(model.predict(x) > 0.5).astype("int32")`, if your model does binary classification (e.g. if it uses a `sigmoid` last-layer activation).
warnings.warn('`model.predict_classes()` is deprecated and '
#mask = "../input/with-and-without-mask/"
plt.figure(figsize=(8,7))
label = {0:"With Mask",1:"Without Mask"}
color_label = {0: (0,255,0),1 : (0,0,255)}
cascade = cv2.CascadeClassifier("frontalface/haarcascade_frontalface_default.xml")
count = 0
i = "/content/with-and-without-mask/masks2.0/masks/test/1/1.jpeg"
frame =cv2.imread(i)
gray = cv2.cvtColor(frame,cv2.COLOR_BGR2GRAY)
faces = cascade.detectMultiScale(gray,1.1,4)
models = [convnet_model, inception_model, mobilenet_model, densenet_model, vgg19_model, efficientnet_model, resnetv2_model]
for model in models:
for x,y,w,h in faces:
face_image = frame[y:y+h,x:x+w]
resize_img = cv2.resize(face_image,(150,150))
normalized = resize_img/255.0
reshape = np.reshape(normalized,(1,150,150,3))
reshape = np.vstack([reshape])
if model == convnet_model:
name = 'ConvNet'
if model == inception_model:
name = 'InceptionV3'
if model == mobilenet_model:
name = 'MobileNet'
if model == densenet_model:
name = 'DenseNet'
if model == vgg19_model:
name = 'VGG19'
if model == efficientnet_model:
name = 'EfficientNet'
if model == resnetv2_model:
name = 'ResNet50V2'
result = model.predict_classes(reshape)
if result == 0:
cv2.rectangle(frame,(x,y),(x+w,y+h),color_label[0],3)
cv2.rectangle(frame,(x,y-50),(x+w,y),color_label[0],-1)
cv2.putText(frame,label[0],(x,y-10),cv2.FONT_HERSHEY_SIMPLEX,1,(255,255,255),2)
frame = cv2.cvtColor(frame,cv2.COLOR_BGR2RGB)
plt.imshow(frame)
elif result == 1:
cv2.rectangle(frame,(x,y),(x+w,y+h),color_label[1],3)
cv2.rectangle(frame,(x,y-50),(x+w,y),color_label[1],-1)
cv2.putText(frame,label[1],(x,y-10),cv2.FONT_HERSHEY_SIMPLEX,1,(255,255,255),2)
frame = cv2.cvtColor(frame,cv2.COLOR_BGR2RGB)
plt.imshow(frame)
plt.title(name)
plt.axis("off")
plt.show()
cv2.destroyAllWindows()
/usr/local/lib/python3.7/dist-packages/keras/engine/sequential.py:450: UserWarning: `model.predict_classes()` is deprecated and will be removed after 2021-01-01. Please use instead:* `np.argmax(model.predict(x), axis=-1)`, if your model does multi-class classification (e.g. if it uses a `softmax` last-layer activation).* `(model.predict(x) > 0.5).astype("int32")`, if your model does binary classification (e.g. if it uses a `sigmoid` last-layer activation).
warnings.warn('`model.predict_classes()` is deprecated and '
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/sequential.py:455: UserWarning: `model.predict_classes()` is deprecated and will be removed after 2021-01-01. Please use instead:* `np.argmax(model.predict(x), axis=-1)`, if your model does multi-class classification (e.g. if it uses a `softmax` last-layer activation).* `(model.predict(x) > 0.5).astype("int32")`, if your model does binary classification (e.g. if it uses a `sigmoid` last-layer activation).
warnings.warn('`model.predict_classes()` is deprecated and '
